Expectation maximization approach to gross error and change point detection

Accuracy of process measurements is critical in process operation and control. However, in reality, miscalibration or malfunctioning of instruments may introduce bias or gross error resulting in abnormal process operation and poor control performance. Timely identification of these biased instruments and rectifying them have a great impact on process control performance. In this paper, two new probabilistic methods based on Expectation Maximization are proposed for detecting biased instruments as well as detecting the abnormal time point. Performances of the proposed EM based algorithms are compared with Bayesian algorithm. Simulation results show the power and efficiency of EM in gross error detection especially when the priors are chosen improperly.